Simulation-Based Evaluations of Reinforcement Learning Algorithms for Autonomous Mobile Robot Path Planning

Abstract

This work aims to evaluate the efficiency of the five fundamental reinforcement learning algorithms including Q-learning, Sarsa, Watkins’s Q(λ), Sarsa(λ), and Dyna-Q, and indicate which one is the most efficient of the five algorithms for the path planning problem of autonomous mobile robots. In the sense of the reinforcement learning algorithms, the Q-learning algorithm is the most popular and seems to be the most effective model-free algorithm for a learning robot. However, our experimental results show that the Dyna-Q algorithm, a method learns from the past model-learning and direct reinforcement learning is particularly efficient for this problem in a large environment of states.